Audience Build Execution — Jun 11, 2026

The data prep step is done. Four ready-to-upload CSVs are sitting in /Users/mo/Downloads/For FB Audiences/output/. This doc walks through what's in them, the Ads Manager upload sequence, and what to do if match rates underperform.

What just happened

The script prep_audience_csvs.py read both PII files Anthony Kaylor delivered, filtered to Military.com Auto + Home campaigns only (using the campaign-ID mapping from his email), renamed columns to Meta's customer-list spec, and deduplicated on email. Result: 4 clean CSVs in /Users/mo/Downloads/For FB Audiences/output/ ready for direct upload.

Output files

File	Use	Rows	Source window
`mil_combined_seed_90d.csv`	Fresh seed for general LAL ladder (warm layer)	112,264	Mar-May 2026, Auto + Home
`mil_auto_seed_90d.csv`	Auto-specific LAL ladder for Auto ALC	91,871	Mar-May 2026, Auto only
`mil_home_seed_90d.csv`	Home-specific LAL ladder for Home ALC	21,777	Mar-May 2026, Home only
`mil_suppression_all_7mo.csv`	Existing-customer suppression list	195,137	Nov 2025 - May 2026, all Mil.com

Data quality check (from the script run)

100% field completeness across email, first name, last name, ZIP, state, country, DOB, birthyear, gender on all 4 outputs. This is the best-case scenario for Meta match rates.
21,042 rows from the raw Mar-May file were skipped (12%) because they had no email. Meta needs email at minimum.
No phone column in the source data. Anthony Kaylor's email explicitly offered to add this from QW if needed — we'll use that escape hatch only if match rates come in below 60%.
Cross-file dedup worked: 82,873 unique emails from the older file (Nov-Feb) that weren't already in the Mar-May file got added to suppression.

Ads Manager upload sequence

Total time: ~60 minutes (mostly waiting for Meta to process).

Step 1 — Upload fresh seed (Combined)

Business Manager → Audiences → Create Audience → Custom Audience → Customer List → Use a list of customers → Next

Use customer lifetime value data: No (we use this for prospecting, not value modeling)
Original data source: Directly from customers
Audience name: Custom_Recruiting-Leads_06-11-2026_Fresh
Description: "Mar-May 2026 QW accepted+sold leads, Auto+Home combined. 112,264 records. Replaces 08-05-2025 seed."
File to upload: /Users/mo/Downloads/For FB Audiences/output/mil_combined_seed_90d.csv
Identifier mapping: Meta should auto-detect the columns (email, fn, ln, etc.). If it asks for manual mapping, confirm each maps to the matching Customer Information field.
Hash: Yes (Meta auto-hashes — check the box if asked)
Click Upload & Create

Meta begins processing. Audience size will show "Calculating..." for 30-60 minutes.

Step 2 — Upload Auto-specific seed (optional but recommended)

Same flow as Step 1.

Audience name: Custom_Auto-Leads_06-11-2026_Fresh
Description: "Mar-May 2026 QW Auto-only leads. 91,871 records. Seed for Auto ALC LAL ladder."
File: /Users/mo/Downloads/For FB Audiences/output/mil_auto_seed_90d.csv

Step 3 — Upload Home-specific seed (optional but recommended)

Audience name: Custom_Home-Leads_06-11-2026_Fresh
Description: "Mar-May 2026 QW Home-only leads. 21,777 records. Seed for Home ALC LAL ladder."
File: /Users/mo/Downloads/For FB Audiences/output/mil_home_seed_90d.csv

Step 4 — Upload suppression list

Same flow.

Audience name: Suppression_Existing-Customers_All-Time_06-11-2026
Description: "All Mil.com Auto+Home customers Nov 2025 - May 2026. 195,137 records. EXCLUDE from every prospecting ad set. Refresh monthly."
File: /Users/mo/Downloads/For FB Audiences/output/mil_suppression_all_7mo.csv

Step 5 — Wait for processing (30-60 min)

Go check on Auto ALC's underutilization in the meantime, or build the Warm Layer ad set (it doesn't depend on these customer-list uploads — see the Audience Structure Spec).

Step 6 — Validate match rates

Once each audience shows "Audience size" instead of "Calculating...", note the size. Match rate is approximate:

Match rate ≈ (Audience size shown by Meta) / (Rows you uploaded)

Audience	Rows uploaded	Target Audience size (60% match)	Target (75% match)
Combined seed	112,264	≥ 67,000	≥ 84,000
Auto seed	91,871	≥ 55,000	≥ 68,000
Home seed	21,777	≥ 13,000	≥ 16,000
Suppression	195,137	≥ 117,000	≥ 146,000

If match rate < 60% on any audience: email Anthony Kaylor accepting his phone-number offer. Draft below.

If 60-80% (likely outcome): proceed to build LAL audiences.

If 80%+: even better — proceed normally.

Step 7 — Build LAL ladders (after seeds finish processing)

Per the Audience Structure Spec, Option C (hybrid: Combined LAL for Warm Layer, vertical-specific LALs for ALC) is the recommended approach.

LAL set 1 — Combined LAL ladder (3 audiences)

Audiences → Create → Lookalike → Source: Custom_Recruiting-Leads_06-11-2026_Fresh

Build three at once: - LAL_US_1pct_Custom-Recruiting-Leads_06-11-2026_Fresh (size 1%, ~2M people) - LAL_US_3pct_Custom-Recruiting-Leads_06-11-2026_Fresh (size 3%, ~6M people) - LAL_US_5pct_Custom-Recruiting-Leads_06-11-2026_Fresh (size 5%, ~10M people)

Location: US. Click Create.

LAL set 2 — Auto-specific LAL ladder

Source: Custom_Auto-Leads_06-11-2026_Fresh

LAL_US_1pct_Auto-Leads_06-11-2026_Fresh
LAL_US_3pct_Auto-Leads_06-11-2026_Fresh
LAL_US_5pct_Auto-Leads_06-11-2026_Fresh

LAL set 3 — Home-specific LAL ladder

Source: Custom_Home-Leads_06-11-2026_Fresh

LAL_US_1pct_Home-Leads_06-11-2026_Fresh
LAL_US_3pct_Home-Leads_06-11-2026_Fresh
LAL_US_5pct_Home-Leads_06-11-2026_Fresh

All LAL builds process in 5-15 minutes typically.

Step 8 — Wire into campaigns (per Audience Structure Spec, with updated LAL assignments)

The Audience Structure Spec said "use new LAL ladder in Auto ALC and Home ALC." With vertical-specific LALs now available, use this assignment instead:

Campaign	Ad Set	Include	Exclude
Auto BAU	Broad (existing — modify)	(no custom audience — broad targeting + Advantage+ ON)	Suppression list
Auto BAU	Custom (existing — modify)	`Custom_Recruiting-Leads_06-11-2026_Fresh` (combined seed)	Suppression list
Auto BAU	NEW Warm Layer	4 page-engagement audiences (FB Likers + Engagers + IG Followers + Engagers)	Suppression list
Auto Max Val	Broad (modify)	(broad + Advantage+)	Suppression list
Auto ALC	Broad 2.O (modify)	Auto-specific LALs: `LAL_US_1pct_Auto-Leads_06-11-2026_Fresh` + 3% + 5%	Suppression list
Home BAU	Broad (modify)	(broad + Advantage+)	Suppression list
Home BAU	Custom (modify)	`Custom_Recruiting-Leads_06-11-2026_Fresh`	Suppression list
Home BAU	Home Insurance (existing)	(keep as is, add suppression)	Suppression list
Home BAU	NEW Warm Layer	4 page-engagement audiences	Suppression list
Home Max Val	Broad (modify)	(broad + Advantage+)	Suppression list
Home ALC	Broad (modify)	Home-specific LALs: `LAL_US_1pct_Home-Leads_06-11-2026_Fresh` + 3% + 5%	Suppression list

Plus the Auto ALC overlap fix from the Audience Structure Spec: Auto ALC's targeting must have Detailed Targeting Expansion OFF and Advantage+ Audience OFF. Auto BAU has both ON. This makes the two campaigns structurally distinct so Meta no longer self-limits.

Step 9 — Cleanup (after 72hr stability check)

Once new LAL ladders are deploying at positive ROAS for 3 days:

Delete the 28 orphaned placeholder LALs (all sized 1,000, INACTIVE — bulk select in Audiences filter)
Delete the old stale LAL ladder: - Lookalike (US, 1%) - Custom Audience_08-05-2025 - Lookalike (US, 3%) - Custom_Recruiting-Leads_08-25 - Lookalike (US, 5%) - Custom_Recruiting-Leads_08-05-2025
Delete the old stale seed: Custom_Recruiting-Leads_08-05-2025

Don't delete before the new structure has been stable for 72 hours — rolling back is harder once the source data is gone.

What's still missing

1. 30-day form-submitter cooldown list

Anthony Kaylor's delivery only includes ACCEPTED + SOLD leads. The Audience Structure Spec also calls for a 30-day form-submitter cooldown — all users who hit the quote form, regardless of QW outcome. This data lives in FunnelFlux / Iron Corp's funnel database, not in QW.

Email draft for Lauren:

Lauren — for the audience cleanup I'm doing in Ads Manager, I have the QW accepted/sold leads (Anthony Kaylor sent those). I'm also looking to build a 30-day form-submitter suppression list — all users who hit the quote form in the last 30 days, whether or not they ended up accepted by QW.

Is that data accessible from FunnelFlux or wherever your funnel submission events land? Same format as Anthony's exports would be ideal (email + fn + ln + zip + dob + state + country + gender, raw — Meta auto-hashes on upload). Last 30 days rolling window. Want to use this to give recent submitters a cooldown before retargeting them again — prevents wasting impressions on users who just submitted.

If easier I can pull it myself once you point me at the right access. Thanks!

2. Phone numbers (only if match rate <60%)

Email draft for Anthony Kaylor (only send if first match rate comes in below 60%):

Hey Anthony — thanks again for the PII exports. The combined seed matched at [X]% which is below the 60% threshold Meta uses for confident audience targeting. Want to take you up on your offer to pull phone numbers from QW and re-export — based on Meta's docs that should lift match rates by 10-15% which would put us comfortably in the deploy-confidently range.

Same campaign ID filters as last time (Auto: C20188, C32405, C32535, C32533, C31885, C31476, C32381, C31701; Home: C20259, C32454, C31886, C31633). Just add a phone column to the existing schema.

No urgency — first match rate is workable for the immediate build, but the phone-augmented version would help us scale the LAL ladder more confidently. Thanks!

3. Recurring refresh cadence

After this build is live, set a calendar reminder to ask Anthony Kaylor for a fresh export every 30 days. Re-run the prep script, re-upload to Meta (Meta's UI has an "Update audience" flow that preserves the audience ID — just replace the data behind it).

The Python script (for reference)

Lives at /Users/mo/Desktop/military.com/deliverables/audit_2026-06-04/prep_audience_csvs.py.

Re-run anytime by running:

python3 /Users/mo/Desktop/military.com/deliverables/audit_2026-06-04/prep_audience_csvs.py

It auto-creates /Users/mo/Downloads/For FB Audiences/output/ if it doesn't exist and overwrites any prior CSVs.

To use it for the next monthly refresh: 1. Drop the new file into /Users/mo/Downloads/For FB Audiences/ (same naming convention or update the script's MAR_MAY / NOV_FEB constants) 2. Re-run the script 3. Re-upload the resulting CSVs to Meta

Verification (after Step 6)

The audience build is working when:

✅ All 4 customer-list audiences show "Audience size" in Ads Manager (not "Calculating")
✅ Match rates are ≥60% (≥75% is great)
✅ Audience sizes (after Meta's match) are roughly: Combined ~70-95K, Auto ~55-75K, Home ~13-18K, Suppression ~120-160K
✅ 3 LAL ladders successfully built (Combined, Auto, Home — 9 LAL audiences total)
✅ Per-campaign include/exclude wired per the matrix above
✅ 28 orphaned placeholder LALs deleted (after stability check)
✅ Auto ALC daily utilization starts rising from 17% over the next 5-7 days

If Auto ALC utilization doesn't move within 4 days of the build, the constraint isn't audience overlap — investigate other gating factors (bid strategy, optimization event volume, learning phase).

ASSUMPTIONS & OPEN QUESTIONS

Vertical-specific LAL approach (Option C): Recommended over the combined-only approach. Costs 15 extra minutes of Ads Manager work for better Meta optimization signal on Auto ALC + Home ALC.
Source data assumed correct: Trusting Anthony Kaylor's campaign-ID mapping. If the IDs include non-Mil.com data anywhere, the suppression list could over-suppress legitimate prospects. Spot-check a few rows from each campaign ID to confirm they look like Mil.com customers.
No phone field = lower match rate ceiling: Without phone, expect match rates in the 55-75% range. Anthony's offer to add phone if rates are low is the escape hatch.
21K records skipped (no email): Acceptable — these wouldn't have matched in Meta anyway. Not worth pursuing.
Recipients-only data: The data is QW accepted+sold leads. We don't have the "form submitter but didn't get accepted by QW" cohort, which is the larger user base. Lauren's 30-day cooldown list would fill this gap.
Conversion_value column: Present in source data but not included in upload (Meta supports value-based custom audiences but the Audience Structure Spec opted for prospecting use case). Reserved for future test if Lauren wants to try value-based optimization separately.

← Back to Military.com portal